Identificação da região de interface para monitoração no transporte de subprodutos de petróleo em polidutos usando radiação gama e redes neurais artificiais

Abstract

This study presents a methodology to identify the interface region of petroleum by-products transported in polyducts using gamma-ray densitometry and an artificial neural network to reduce the volume of contaminated products. The mathematical simulation of the measurement geometry is composed of a 662 keV (137Cs) gamma-ray source and scintillation detectors. The source is collimated to obtain a narrow radiation beam, and two 1¼×¾" NaI(Tl) detectors were used, one to measure the transmitted beam and the other to measure the scattered beam. The response function of a real detector was experimentally validated and provided more realistic data for mathematical simulation. Several radii of this steel duct were investigated ranging from 4 to 10 inches. Static models were developed using the MCNP6 code, for a stratified flow regime. Fluids found in the oil industry, gasoline, kerosene, fuel oil and glycerol were used to compose various combinations of purity levels. In this way, the spectra recorded in both detectors were used as input data for training and evaluation of artificial neural networks. Furthermore, an intelligent system based on two neural networks (Classifier and Predictor) was developed to increase the accuracy of the results. The proposed methodology has the potential to identify the interface region presenting 1% accuracy as the degree of impurity. These results were evaluated with the mean relative error and root mean squared error metrics and presented, respectively, 1.12% and 0.204 for all investigated patterns.